ESTRO 2025 - Abstract Book

S1346

Clinical - Lung

ESTRO 2025

Conclusion: AI autosegmentation significantly reduces the time needed for ATS nodal contouring when undertaken across multiple 4D datasets in lung cancer radiotherapy while maintaining accuracy comparable to expert manual contouring. Raw-AI agrees well with expert manual delineation. However, screening for large deviations in raw AI is important. Consistency and efficiency will be evaluated on a larger database.

Keywords: ATS Nodes, AI segmentation, 4DCT

References: 1. Nestle U et al, ESTRO ACROP Guidelines for target Volume definition in the treatment of locally advanced non-small cell lung cancer , Radiother Oncol 2018, Apr; 127(1) : 1-5 2. Le Pechous C et al, ESTRO ACROP Guidelines for target Volume definition in the treatment of locally advanced small cell lung cancer, Radiother 2020, Nov: 152:89-95

2243

Digital Poster Fractal dimension and lacunarity as predictive radiomic features for relapse in pulmonary nodules Maria Pagola 1 , Miren Josune Urien 2 , Eva Saenz de Urturi 1 , Alai Goñi 1 , Belen De Paula 1 , Usoa Iceta 1 , David Ortiz de Urbina 1 , Nuria Bulto 1 , Julian Minguez 1 , Carlos Blanco 1 , Mikel Eguiguren 1 , Elena Guimón 1 , Intza Uranga 1 , Daniel Alberto Roura 1 , Leyre Gonzalez 1 , Ane Mugica 1 , Amaya Sanchez 1 , Maider Campo 1 , Beraldo Martinez 1 , Ane Otaegui 1 , Xabier Gurutzeaga 1 , Ane Dehesa 1 , Ainhoa Diez 1 , Sebastian Luevano 1 , Arrate Querejeta 1 1 Radiation Oncology, Onkologikoa - UGC Oncología, Gipuzkoa, Spain. 2 Faculty of Engineering, Mondragon University, Gipuzkoa, Spain Purpose/Objective: Radiomic features such as fractal dimension and lacunarity have shown promising results in tumor analysis, characterizing the structural complexity and internal heterogeneity of pulmonary nodules in CT images. However, select those providing relevant information without redundancy of the large number of available radiomic variables is necessary. This preliminary study investigates the predictive potential of fractal-dimension and lacunarity for early relapse in patients with pulmonary nodules treated with SBRT. histologically confirmed or with high suspicion of lung cancer, treated with SBRT-Tomotherapy from 2013 to 2019. Tumor geometry and fractality related radiomic variables were extracted from CT images within the PET/CT studies. Four representative variables were chosen by combining correlation studies and classification tree (CRT) analysis to minimize redundancy and optimize predictive performance: median 2D fractal dimension across tumor slices, lacunarity in the slice with the largest tumor cross-section, the tumor volume-to-minimum bounding volume ratio, and the tumor surface area-to-volume ratio. Fractal dimension captures the structural complexity of the tumor, while grayscale lacunarity reflects textural heterogeneity, linked to density variability and tumor growth patterns. A predictive model was built using CRT with cross-validation, incorporating the selected variables to predict tumor relapse local (LR), pulmonary (P), locoregional (R), or distant (D). Results: Eighty-nine patients with segmented nodules were analyzed. At 12 months, relapse patterns were 6LR, 9P, 12R, and 3D. The model achieved a ROC AUC of 0.955, demonstrating excellent predictive performance, with a sensitivity of 93.8%, specificity of 86.3%, and overall accuracy of 89.3%. High lacunarity correlated with greater textural heterogeneity and relapse risk. Tumors with higher 2D fractal dimension values exhibited higher relapse Material/Methods: We retrospectively analyzed pre-treatment PET/CT images of patients with solitary pulmonary nodules,